Low-Complexity Classification Algorithm to Identify Drivers’ Stress Using Electrodermal Activity (EDA) Measurements

We present a system where a simple and low-complexity classification algorithm is used to identify the stress of a person while driving a car, using EDA Skin Potential Response (SPR) measurements. An adaptive filter, which takes the steering wheel signal as a reference signal, is used to remove the motion artifacts that appear in the recorded SPR signal as a consequence of hand movements introduced by steering the wheel and by vibrations. Statistical features are then extracted from the resulting signal, which should well represent the emotional and stress components of the SPR signal. These features are given as an input to a Support Vector Machine (SVM) classifier in order to detect the existence of stress in a given time interval. Data are collected from tests on different subjects, carried out in a scenario where stress is induced at random moments through sudden sounds, with a metronome frequency ticking sound that gives the pace for the steering wheel movement. An accuracy of 87.40% is obtained when we consider both the stress triggers and the metronome frequency change as stress-inducing events for the subjects. We then utilize our classification system with real data confirming the good performance of our system.